Crowd counting with WiFi sensing based on iterative attentional feature fusion

—Crowd counting has great appeal for a variety of applications, such as public transportation, disaster management and building automation. Recently, WiFi-based crowd counting has gained dominance due to its ubiquitous and non-invasive advantages. However, current WiFi-based crowd counting systems have a limitation in that they do not consider the effect of dynamic crowds and static crowds on crowd counting. In contrast to previous studies, this paper investigates the effect of crowds in different states on crowd counting performance, and proposes a WiFi-based multi-state crowd counting system, which can not only count dynamic or static crowds, but also count joint dynamic and static crowds. By analyzing the effect of crowd states on the signal, we demonstrate that the channel state information (CSI) subcarrier distribution can indicate the count of crowds in different states. To this end, we adopt an iterative attentional feature fusion (IAFF) which allows for the fusion of amplitude and phase information from multiple antennas and adaptively assigns weights to amplitude and phase on multiple subcarriers, thus enabling the counting of crowds in various states. The experimental results show that the system has recognition accuracy of 99.38 % for static crowds, 95.94 % for dynamic crowds, and 97.57 % for joint dynamic and static crowds.